CVE-2026-68258

What it actually does

Missing bounds checks on queue type and mqd_size values supplied in private data during CRIU restore in the AMD KFD driver allow invalid or out-of-range inputs to be accepted and processed without validation.

Why it matters to the business

Systems using AMD GPUs may experience kernel instability, crashes, or incorrect queue behavior during checkpoint and restore operations, affecting reliability in container, VM migration, or HPC workloads.

How an attack unfolds

An attacker supplies crafted CRIU restore data containing a queue type at or above the maximum or an mqd_size that does not match the queue type; the driver previously processed these values without verification, potentially resulting in invalid internal state or memory access issues.

How to fix it

Update the Linux kernel to a version containing the bounds checks for KFD_QUEUE_TYPE_MAX and the mqd_size_from_queue_type validation function.

How to confirm the fix worked

Confirm the presence of the added KFD_QUEUE_TYPE_MAX constant and mqd_size_from_queue_type function in the drm/amdkfd CRIU restore path; test restore operations with boundary and invalid values to ensure rejection.

Typical attack path

1. Gain ability to initiate or influence CRIU restore on a target system with AMD KFD
2. Craft restore payload containing out-of-bounds queue type or mismatched mqd_size
3. Trigger the CRIU restore operation
4. Observe or leverage resulting invalid driver state

How to detect it

Examine kernel logs for unexpected queue type or mqd size values during CRIU restore attempts; no public exploitation indicators are known.

What your security team should do

• Apply the upstream kernel fix that adds the missing bounds validation
• Prioritize updates on systems performing CRIU operations with AMD GPUs
• Review kernel logs for CRIU restore errors after patching